Search papers, labs, and topics across Lattice.
This paper introduces GDPevo, a novel benchmark designed to evaluate agent self-evolution in economically relevant business tasks by utilizing a fully automated data pipeline. The benchmark employs rule hybridization to ensure that test-time performance improvements can be directly attributed to prior training experiences, addressing limitations of existing evaluation methods. Results show that self-evolution enhances accuracy by up to 16.44 percentage points across various agents, though performance still lags significantly behind the oracle ceiling, highlighting the need for further advancements in agent self-evolution capabilities.
Self-evolution boosts agent accuracy by over 16% in real-world business tasks, yet current models still fall short of optimal performance.
Agent self-evolution updates an agent's persistent state from prior experience and reuses it to solve related tasks more effectively. Evaluating self-evolution is difficult: existing benchmarks provide limited coverage of economically valuable task domains, do not always design training and test tasks such that test-time gains can be attributed to training experience, and remain vulnerable to data contamination. We present GDPevo, an evolution-native benchmark grounded in GDP-related enterprise workflows, together with the fully automated data pipeline that generates it. Its core mechanism, rule hybridization, decomposes each enterprise workflow into atomic business rules, distributes subsets of these rules across training tasks, and recombines them in held-out test tasks so that test-time gains are attributable. GDPevo spans CRM, ERP, finance, healthcare, legal, and data-centric workflows. Its V1 release contains 120 tasks in 12 groups, with five training and five held-out test tasks per group. Full automation enables the pipeline to expand the suite to 240 tasks in 24 groups (V2) within two days, providing a practical response to contamination. Using GDPevo, we evaluate four agents, each comprising a harness and a model, under four supervision types. Self-evolution consistently improves held-out accuracy by up to 16.44 percentage points. But the best evolved agents remain far below the fully informed oracle ceiling of 91.6%, indicating that the self-evolution ability of current agents remains far from fully realized. We publicly release the pipeline, benchmark, and full evaluation results at https://github.com/Prism-Shadow/GDPevo.
